As head of Vaccine Manufacturing at Sanofi, a global pharmaceutical and healthcare company, Mary Oates has a major role in the company’s mission to “chase the miracles of science to improve people’s lives.” One innovation that is helping to accelerate Sanofi’s delivery of life-saving medicines? The virtual twin. Compass asked Oates how Sanofi is using these scientifically accurate visualizations of its vaccine manufacturing plants and processes to test, refine and implement improvements that enhance quality, lower costs, improve worker safety and ensure ample drug supply.
How does virtual twin technology benefit vaccine manufacturing operations like yours, and can you give me a Sanofi-specific example?
Virtual twins enable us to build, model, learn and improve in the digital world, without suffering any consequences if something we try doesn’t work as planned. By testing our plan in the twin, we can know before we take an action that it will accomplish our goal, or if we need to make a new plan. For example, in one of our plants, we knew that production was going to increase and there was already a bottleneck in the washing area. This is a very simple environment: the dirty formulation tanks are brought in, get disassembled, cleaned, reassembled and sterilized, and then they are sent back out to production.
While the steps are pretty simple, they involve a tremendous number of interactions between all the different parts that comprise the washing area. Our goal was to determine if we could bring in additional manufacturing volume without having to expand that area. So we decided to build a virtual twin to see if we could find the answer to that question.
What data did you use to build the twin?
We didn’t have any data-collection sensors in that part of the plant, so we had to collect the data manually. We determined the availability of qualified operators and equipment, and how long we needed the equipment for each step in the process.
We also looked at the time required for routine and non-routine maintenance. For example, how often did the equipment break down and go out of service? We looked at SOPs [standard operating procedures]; what steps did the operators have to follow and in what order? We also determined unwritten ways of working, things that are not listed in the SOPs.
We talked to the operators, asking them how they work: how they manage lunch, how they manage breaks, how they organize the flow – things that weren’t written down. We also observed the area in operation, and we documented everything that happened in each area. We did that for a period of three months, because we wanted to make sure we had enough data to say: this is how the washing area really operates on a day-to-day basis.
From that point, what was your process?
After we collected all the information, which was rather labor intensive, we entered the data into the virtual twin software, and we modeled the entire process. And then, using the advantage of the twin, we adjusted the parameters. We adjusted them one by one and we adjusted them collectively to see what difference they would make, and to evaluate if we could increase the capacity of the washing area.
We also tested the gut instincts that our operators had about what they thought would make things better. Surprisingly, more often than not, their gut instinct turned out to be incorrect.
The twin showed us, for example, that adding more washing stations or another autoclave only moved the bottleneck further down the process. Instead, we discovered we had to optimize in a range of areas to increase the overall capacity of the washing area. So we learned that, yes, we can accommodate the additional manufacturing volume without increasing resources inside the washing area, simply by making the improvements we identified. We are in the implementation phase of this now.
What else did you learn from this project?
The beauty of a virtual twin is that you can try something and fail, again and again, and it doesn’t cost you anything in terms of tying up the physical area or expenditure of resources, apart from building the twin. A couple of other things that we learned from this exercise relate to the expression “garbage in, garbage out.” We have to make sure that the data that goes in is accurate and up to date. And, because there is no direct connection in this case between the physical world and the digital world, we need to manually update that virtual twin to reflect changes in what we are doing in the physical washing area.
Are you using virtual twins in other ways as well?
Let me share a more complex example in Sanofi’s forthcoming EVolutive Facility (EVF), where we plan to use virtual twins throughout the entire vaccine lifecycle. We are currently building the EVFs in France and in Singapore, which will be essentially identical. During the building phase, we are creating a virtual twin for the design review. We will input the entire facility layout to create an accurate 3D layout. We are going to identify the locations of every single object in the space; for example, where each cable, plug and socket will be.
Once we have that, we will layer in where the equipment will be located. So whether it’s single- use equipment or stationary equipment, we will know where every single piece will be. Once we have that virtual twin, we will be able to evaluate the space for any potential issues or clashes. For example, will a wall cover a pipe that we will need to access in the future?
We’ll also be able to make sure that the operations team has a complete view of the space. So it’s not just the engineers building it that have the visibility; now the operators will be able to look as well to see the flow of people, materials and waste through this facility to make sure that everything will work as is intended. There’s certainly some software today outside of virtual twins that allows you to do this, but it’s the building blocks that make the virtual twins so incredibly useful.
Can you elaborate on these “building blocks?”
The processes that we will use in the EVFs will be developed by R&D in another facility, then transferred into the EVF. Once we understand what the processes coming to us from R&D are, we can then layer them into the virtual twin. This will then allow us to understand exactly how each process will work in the facility: how the equipment will run, what we anticipate the results will be, because we’ll be able to simulate the process in the EVF twin without ever physically going into a facility.
That will allow us to make any modifications that we think are necessary in terms of the process, given the physical space and the equipment that we have available to us. Then we can transfer these processes into the physical EVF.
Because we will have sensors on all of the equipment and across the environment, all of that information will flow into the twin. Through all this data coming in, we will be able to continuously learn and compare results to the original process transferred to us from R&D. That allows us to be fairly confident that the time required from initial introduction of the process to when we are ready to do validation lots of vaccine will be approximately 50% less than with conventional processes. So, through the use of a virtual twin, we’ll be able to bring our new vaccines to market much quicker than we have historically.